WEB COMPUTING
Editor: Geoffrey Fox, fox@csit.fsu.edu
PEER-TO-PEER NETWORKS
By Geoffrey Fox
W ILL PEER-TO-PEER COMPUTING BE THE NEXT KILLER
INTERNET APPLICATION? LIKE MOST OVER-HYPED
CONCEPTS, P2P IS LOOSELY DEFINED AND COVERS A SET OF
RATHER DISPARATE IDEAS. PERHAPS THE ONLY COMMON THEME IS
a client-oriented view of the world;
you might think of P2P as “Power
to the People.” The servers are sub-
servient to the clients, which do most
of the work.
P2P has several important technol-
ogy challenges and applications, vary-
ing from the sublime to the ridiculous.
This column presents a quick overview
and suggests some emerging research
areas and opportunities.
Napster
Napster (www.napster.com) is the
most well-known and popular P2P sys-
tem. Shawn Fanning developed the
original application and service in Jan-
uary 1999 while a freshman at North-
eastern University. Napster lets any
client advertise the MP3 files stored on
its disk and download MP3 files from
other clients connected to the Napster
server network. It is said that Fanning
was taking a computer programming
course at Northeastern but had to buy
a programming book to build Napster.
Like most good ideas, Napster was de-
signed to solve a real need—in this case
to enable Fanning, a musician, to share
his music with his friends on campus.
The system has become staggeringly
popular. A legal opinion from last sum-
mer stated that users were sharing ap-
MAY/JUNE 2001
proximately 10,000 music files per sec-
ond using Napster, more than 100 users
tried to connect to the system every sec-
ond, and there would be 75 million
Napster users by the end of 2000
(http://news.cnet.com/News/Pages/
Special/Napster/napster_patel.html).
Napster has other typical P2P ser-
vices—instant messaging, chat rooms,
buddy lists, and information about to-
day’s popular music—but the key fea-
ture is the ability to share files between
any Internet-connected consenting
clients. This is roughly the Web ver-
sion of NFS (Network File System) fa-
miliar from traditional computing en-
vironments.
Another feature is its handling of
MP3 files, which are important as a
popular digital encoding format for au-
dio files. It is straightforward to “rip”
(copy) files off an audio CD and look
up key metadata (artist, title, and so on)
in the CDDB database on the Web
(www.gracenote.com). You can store
and access the audio and metadata as a
single unit. Although the system uses a
server to establish the initial connec-
tion, it transfers files efficiently, di-
rectly from client to client. This is an
improvement over most NFSs where
using distributed files is not easy (ex-
cept possibly for the originator), be-
cause usually all you have is a filename.
The added value of metadata for files
lies at the heart of the Semantic Web,
a vision from the W3C Web Consor-
tium related to P2P (www.w3.org/
2001/sw).
Although the two concepts of Web-
based NFS and metadata-enhanced
files are fundamental and broadly ap-
plicable, Napster is currently contro-
versial and under siege because the
audio files are typically copyrighted.
The company’s legal problems are a
feature of the particular content, but
the technology is long-lasting and in
my opinion uncontroversial—it’s an es-
sential P2P capability.
Some 200 Napster clones are avail-
able to support this area (www.
ultimateresourcesite.com/napster/
main.htm). Currently the most popu-
lar is Imesh (www.imesh.com), which
has some two million users and can
share any type of file. Some of the
best-known file-sharing systems are
MojoNation (www.mojonation.net),
Freenet (http://freenet.sourceforge.
net), and Gnutella (http://gnutella.
wego.com). These are not server-
based like Napster but rather support
waves of software agents expressing
resource availability and interest prop-
agating among an informal, dynamic
network of peers.
Other P2P features
So far, we’ve examined some basic
P2P services—file registration, access,
and search. We can also categorize
P2P systems in other ways, including
distributed computing, collaboration,
75
 WEB COMPUTING
and core technologies. Let’s review
these three areas.
Distributing computing. Distrib-
uted-computing P2P applications are
well illustrated in the CiSE article
“Distributed Projects Tackle Protein
Mystery” by Keri Schreiner in the first
issue of 2001. Schreiner discusses
the use of millions of Internet clients
to analyze data looking for extra-
terrestrial life (SETI@home, http://
setiathome.ssl.berkeley.edu) and the
newer project examining protein fold-
ing (Folding@home, www.stanford.
edu/group/pandegroup/Cosm). This
kind of distributed-computing solution
divides an application into a huge num-
ber of essentially independent compu-
tations plus a central server system that
doles out separate work chunks to each
participating client. In the parallel-
computing community, these problems
are called pleasingly or embarrassingly
parallel. I include this approach in the
P2P category because the computing is
peer-based, even though it transfers
files through peer–server communica-
tion (unlike the largely pure client–
client model of Gnutella and Napster).
SETI@home and Folding@home are
elegantly implemented as screen savers
that you download.
Other projects of this type include
United Devices (www.ud.com/home.
htm based on SETI@home), Ap-
pliedMeta (based on the University
of Virginia Legion project, www.
appliedmeta.com), Parabon Compu-
tation (www.parabon.com), Condor
(from Wisconsin www.cs.wisc.edu/
condor), and Entropia (www.entropia.
com). Other applications for this type
of system include financial modeling,
bioinformatics, Web server perfor-
mance testing, and the scheduling of
different jobs to use idle time on a net-
work of workstations. Ian Foster has
76
given a more detailed review of these
activities at www.nature.com/nature/
webmatters/grid/grid.html and relates
them to computational grids (www.
gridforum.org).
Collaboration. Collaborative sys-
tems form a rather different type of P2P
network: a community of clients work-
ing together and sharing different In-
ternet resources. Instant-messenger
(IM) communication and chat rooms, in
all their various forms, are the most
common examples in this arena. Here,
participating clients exchange messages
with each other. Unlike the file-sharing
case, a user would typically multicast the
same message to multiple clients at the
same time. The best architecture for
this is still under active research. In fact,
I work in this area (see the Garnet Col-
laboration System at http://aspen.csit.
fsu.edu/collabtools). Groove Networks
(www.groove.net), founded by the cre-
ator of Lotus Notes, is the best-known
P2P collaboration project; it uses relay
servers to implement the P2P multi-
cast. Collaboration systems form a P2P
“illusion” using a static or dynamic
suite of servers to optimally route mes-
sages. When the clients are scattered
around the world, the relaying servers
would perhaps be in the “middle” of
the Web; when a group of clients are
clustered, their relay would be “on the
edge” and perhaps dynamically created
on a peer machine of this cluster. Typ-
ically you also need some sort of server
to establish the initial session and man-
age the permanent state. So this type of
P2P application gives a rich mix of true
peers and servers.
In addition to IM, chat, and email,
collaboration systems offer shared re-
sources such as white boards, shared doc-
uments, and audio-videoconferencing.
The HearMe system is a nice example
of the P2P illusion (www.hearme.com).
A central server manages digital audio
conferences with a general mix of phone
and pure Internet audio. As their tech-
nology advances, they should move to
the Groove and Garnet models with dy-
namic relay servers positioned through-
out the Web. All forms of collaboration
use some type of messaging, with the
message (typically called an event) car-
rying a variety of content including the
IM text, pixel changes to record a
changed shared display (frame buffer),
or digital audio packets. XML is the
natural way of encoding such messages,
and the open source Instant Messenger
Jabber (www.jabber.org) provides a
clean framework of this kind. Several
Napster-like systems have based their
service on IM technology; Aimster
(www.aimster.com) is one of the best
known. OpenCola (www.opencola.com)
has a general XML framework to sup-
port P2P systems.
Core technologies. Core tech-
nologies or services include P2P man-
agement, messaging, security, and client
grouping, as well as the file (or more
generally object) registration, discovery,
and access capabilities mentioned ear-
lier for Napster. We must develop these
core capabilities and then define com-
munity standards. Then they can inter-
operate, resulting in bigger and better
P2P systems. Sun Microsystems has
two important technology projects. Jini
(www.sun.com/jini) deserves a column
of its own; it has a beautiful, simple
model for dynamic self-defining ob-
jects. Like Napster peers, these objects
register with distributed servers so that
other peers can discover and access
them. JXTA (from juxtaposition, www.
openp2p.com/pub/a/p2p/2001/02/15/
joy_keynote.html) is a new project from
Sun’s chief scientist Bill Joy that aims at
core P2P capabilities, including peer
grouping and security.
COMPUTING IN SCIENCE & ENGINEERING

Besides the messaging services needed
to implement collaborative P2P systems,
there is also a Java message service called
JMS (http://java.sun.com/products/jms)
that provides the core publish–subscribe
mechanism on which most P2P services
are built. This needs some upgrading to
join the P2P revolution; Sun should add
XML, a more dynamic matching para-
digm (of collaborating peers), and sup-
port for relay servers.
I expect research and commercial ex-
perience to identify more base services
as we better understand the common
needs of P2P systems. Resource man-
agement in such a network must be
an important challenge. It is our
Nirvana—the Web operating system.
Maybe society can live in a hodge-
podge of unstructured knowledge
swept back and forth by armies of
Gnutella agents, but this will not do
for the Enterprise P2P (named by
O’Reilly and the Gartner Group,
www.oreillynet.com/pub/d/547)
needed by Fortune 500 organizations.
Here we will need to manage structured
information within a dynamic P2P
grouping.
The right approach is to generalize
Napster and Jini, ensuring that all ob-
jects are tied to metadata (possibly in a
separate record) that define their dis-
covery, rendering, access, and sharing
characteristics. One homely example is
family photos; usually these are indeed
a melange of folders haphazardly
stuffed in shoeboxes. This gets much
worse for a community event recorded
in shoeboxes across the nation. With
the proper metadata and Enterprise
P2P support, such photos could be
nicely organized and presumably of
greater value.
The future
Many interesting ideas are being
explored. As an example, breaking
MAY/JUNE 2001
shared files into many parts could
increase bandwidth (parallel I/O) and
security (no one site could access files
without cooperation from its peers).
This type of technology is contro-
versial because it makes censorship
very hard. MojoNation has a load-
balancing and scheduling algorithm in
the form of micro payments to reward
those who contribute most to the
community. Gnutella, a family of re-
lated products, is usually described as
a P2P search engine, because its inter-
face is more like a search engine than
a Web file system.
There is one more important char-
acteristic of P2P networks: client het-
erogeneity. Hand-held devices, mobile
phones, special interfaces for those
having physical handicaps, as well as
basic desktops can participate in a sin-
gle P2P session. This requires that
each peer be able to render copied files
and shared objects differently. Careful
design of the XML metadata for both
clients and display devices makes this
quite possible and ripe for research.
There are lots of good research top-
ics and obviously lots of business op-
portunities. P2P networks, as part of
the next wave of the Web, are intellec-
tually challenging to design and so-
cially and intellectually rewarding to
use. They can and will unite us all.
Y ou can find general discussions
on P2P technology at two good
Web sites: www.openp2p.com from the
O’Reilly group and www.peer-to-
peerwg.org from an industry working
group originally initiated by Intel. I
also recommend a remarkable book,
Peer-to-Peer: Harnessing the Power of
Disruptive Technologies by Andrew
Oram, Nelson Minar, Clay Shirky, and
Tim O’Reilly (O’Reilly & Associates,
2001).
How to
Reach CiSE
Writers
For detailed information on
submitting articles, write to
cise@computer.org or visit
computer.org/cise/edguide.htm.
Letters to the Editors
Send letters to
Jenny Ferrero, Contact Editor
jferrero@computer.org
Please provide an email address or
daytime phone number with your
letter.
On the Web
Access computer.org/cise or
ojps.aip.org/cise for information
about CiSE.
Subscription Change of Address
(IEEE/CS)
Send change-of-address requests for
magazine subscriptions to address.
change@ieee.org. Be sure to spec-
ify CiSE.
Subscription Change of Address
(AIP)
Send general subscription and
refund inquiries to subs@aip.org.
Subscribe
Visit ojps.aip.org/cise/subscrib.html
or computer.org/subscribe.
Missing or Damaged Copies
If you are missing an issue or you
received a damaged copy, contact
membership@computer.org.
Reprints of Articles
For price information or to order
reprints, send e-mail to cise@
computer.org or fax +1 714 821
4010.
Reprint Permission
To obtain permission to reprint an
article, contact William Hagen,
IEEE Copyrights and Trademarks
Manager, at whagen@ieee.org.